Difference between revisions of "The Sniper Multi-Core Simulator"

From Sniper
Jump to navigation Jump to search
Line 19: Line 19:
 
[[Paper:Sc2011Carlson|Sniper]] is
 
[[Paper:Sc2011Carlson|Sniper]] is
 
a next generation parallel, high-speed and accurate
 
a next generation parallel, high-speed and accurate
x86-64 simulator.  This multi-core simulator is
+
x86 simulator.  This multi-core simulator is
 
based on the [[Interval Simulation|interval core model]] and the [http://groups.csail.mit.edu/carbon/graphite Graphite] simulation
 
based on the [[Interval Simulation|interval core model]] and the [http://groups.csail.mit.edu/carbon/graphite Graphite] simulation
 
infrastructure, allowing for fast and accurate simulation and for
 
infrastructure, allowing for fast and accurate simulation and for
 
trading off simulation speed for accuracy to allow a range of flexible
 
trading off simulation speed for accuracy to allow a range of flexible
simulation options when exploring different multi-core architectures.
+
simulation options when exploring different homogeneous and heterogeneous multi-core architectures.
Using this methodology, we are able to achieve good
 
accuracy against hardware for 16-thread applications.
 
  
The Sniper simulator allows one to perform timing simulations for
+
The Sniper simulator allows one to perform timing simulations for both multi-program workloads
multi-threaded, shared-memory applications with 10s to 100+ cores,
+
and multi-threaded, shared-memory applications with 10s to 100+ cores,
 
at a high speed when compared to existing simulators.  The main
 
at a high speed when compared to existing simulators.  The main
feature of the simulator is its core model which is based on the [[Interval Simulation|interval core model]],
+
feature of the simulator is its core model which is based on [[Interval Simulation|interval simulation]],
a fast mechanistic core model. The interval model raises the level
+
a fast mechanistic core model. Interval simulation raises the level
 
of abstraction in architectural simulation which allows for faster
 
of abstraction in architectural simulation which allows for faster
 
simulator development and evaluation times; it does so by 'jumping'
 
simulator development and evaluation times; it does so by 'jumping'
between miss events, called intervals. On recent multi-core hardware,
+
between miss events, called intervals.
we see speeds of up to 2 MIPS
+
Sniper has been validated against multi-socket Intel Core2 and Nehalem systems
for SPLASH-2 benchmarks, and almost 3 MIPS for SPEC OMP benchmarks.
+
and provides average performance prediction errors within 25%
 +
at a simulation speed of up to several MIPS.
  
 
This simulator, and the interval core model, is useful for uncore
 
This simulator, and the interval core model, is useful for uncore
 
and system-level studies that require more detail than the typical
 
and system-level studies that require more detail than the typical
one-IPC models. As an added benefit, the interval core model allows
+
one-IPC models, but for which cycle-accurate
the generation of CPI stacks, which shows the number of cycles lost
+
simulators are too slow to allow workloads of meaningful sizes to be simulated.
 +
As an added benefit, the interval core model allows
 +
the generation of CPI stacks, which show the number of cycles lost
 
due to different characteristics of the system, like the cache
 
due to different characteristics of the system, like the cache
hierarchy or branch predictor, to be easily understood.
+
hierarchy or branch predictor, and lead to a better understanding of each component's
 +
effect on total system performance. This extends the use for Sniper
 +
to application characterization and hardware/software co-design.
  
 
=== Features ===
 
=== Features ===
Line 53: Line 56:
 
* [[Interval Simulation|Interval core model]]
 
* [[Interval Simulation|Interval core model]]
 
* [[CPI Stacks]] to gain insight into lost cycles
 
* [[CPI Stacks]] to gain insight into lost cycles
* Parallel, multithreaded simulator
+
* Parallel, multi-threaded simulator
* Multithreaded application support, x86-64, SSE2
+
* Multi-program and multi-threaded application support, x86 and x86-64, SSE2
 
* Validated against the Intel Core 2 microarchitecture (See the [[Frequently Asked Questions|FAQ]] for details)
 
* Validated against the Intel Core 2 microarchitecture (See the [[Frequently Asked Questions|FAQ]] for details)
 
* Full DVFS support
 
* Full DVFS support
 
* Shared and private caches
 
* Shared and private caches
* [[Heterogeneous Configuration|Heterogeneous core configuration support]]
+
* [[Heterogeneous Configuration|Heterogeneous configuration support]]
 
* [http://dx.doi.org/10.1109/ISPASS.2009.4919652 Modern branch predictor]
 
* [http://dx.doi.org/10.1109/ISPASS.2009.4919652 Modern branch predictor]
 
* Supports parallel applications using pthreads, OpenMP, TBB, OpenCL
 
* Supports parallel applications using pthreads, OpenMP, TBB, OpenCL

Revision as of 11:14, 9 June 2012

May 16th: Sniper 3.0 released

June 9th: Sniper tutorial at ISCA'12

Sniper is a next generation parallel, high-speed and accurate x86 simulator. This multi-core simulator is based on the interval core model and the Graphite simulation infrastructure, allowing for fast and accurate simulation and for trading off simulation speed for accuracy to allow a range of flexible simulation options when exploring different homogeneous and heterogeneous multi-core architectures.

The Sniper simulator allows one to perform timing simulations for both multi-program workloads and multi-threaded, shared-memory applications with 10s to 100+ cores, at a high speed when compared to existing simulators. The main feature of the simulator is its core model which is based on interval simulation, a fast mechanistic core model. Interval simulation raises the level of abstraction in architectural simulation which allows for faster simulator development and evaluation times; it does so by 'jumping' between miss events, called intervals. Sniper has been validated against multi-socket Intel Core2 and Nehalem systems and provides average performance prediction errors within 25% at a simulation speed of up to several MIPS.

This simulator, and the interval core model, is useful for uncore and system-level studies that require more detail than the typical one-IPC models, but for which cycle-accurate simulators are too slow to allow workloads of meaningful sizes to be simulated. As an added benefit, the interval core model allows the generation of CPI stacks, which show the number of cycles lost due to different characteristics of the system, like the cache hierarchy or branch predictor, and lead to a better understanding of each component's effect on total system performance. This extends the use for Sniper to application characterization and hardware/software co-design.

Features

cpi-splash2-fft-lg.png

In addition to the main features mentioned above, we have updated the base simulation infrastructure to allow for simulating a larger set of workloads on more recent simulated hardware. Here is the full set of some of the recently added features:

  • Interval core model
  • CPI Stacks to gain insight into lost cycles
  • Parallel, multi-threaded simulator
  • Multi-program and multi-threaded application support, x86 and x86-64, SSE2
  • Validated against the Intel Core 2 microarchitecture (See the FAQ for details)
  • Full DVFS support
  • Shared and private caches
  • Heterogeneous configuration support
  • Modern branch predictor
  • Supports parallel applications using pthreads, OpenMP, TBB, OpenCL
  • Runs SPLASH-2, Rodinia, SPEC OMP and most of PARSEC (See our integrated benchmarks quick-start guide)
  • SimAPI and Python interfaces for monitoring and controlling the simulator's behavior at runtime
  • Single-option debugging of simulator or the application itself
  • Modern Linux-OS support (Redhat EL 5,6/Debian Lenny+/Ubuntu 10.04-12.04+/etc.)
  • Open source software, licensed under the MIT and the Interval Academic License
  • Additional features

You can find additional information on the simulator and its components in our SC'11 paper.

Getting started

Discussion list

Subscribe to the Snipersim mailing list
Email:

Please send any questions or comments to: snipersim [at] googlegroups [dot] com . You can also visit our Google Groups page to subscribe to the list and search through the archive of previous messages.

Team Members

More information

<meta name="keywords" content="Sniper Simulator,Sniper,simulator,computer architecture,multicore,multi core,multi-core,manycore,many core,many-core,micro-architecture,microarchitecture,ExaScience Lab,exascience,exascale,SC11,supercomputers,super-computing,computers,Ghent University,Universiteit Gent"/>